Low-Power Hardware Implementation of Least-Mean-Square Adaptive Filters Using Approximate Arithmetic

Adaptive filters based on least-mean-square (LMS) algorithm are used in several applications in virtue of their good steady-state performance, numerical stability, and acceptable computational complexity. The hardware implementation of LMS filters requires a massive number of multipliers that significantly impact on the power consumption. Approximate computing, a design technique that trades off computation accuracy for better electrical performance, is a way to improve the energy efficiency of LMS filters. In this paper, we implement state-of-the-art approximate multipliers and evaluate their impact on the performance of the LMS algorithm. Moreover, a novel approximate multiplier, whose accuracy can be tuned at design time to better adapt to the application scenario, is proposed. Implementation results in 28-nm CMOS technology allow us to investigate the power versus quality trade-off of the considered LMS approximate filters in two different applications.